当前在线人数16017
Well, what is AI? - 未名空间精华区
首页 - 版面精华区 - 新闻中心精华区 - 科技新闻版精华区 - 精华区文章阅读 首页
未名交友
[更多]
[更多]
Well, what is AI?

发信人: magicfat (魔法胖子~missing my Friend), 信区: ITnews
标  题: Well, what is AI?
发信站: The unknown SPACE (Mon Jan 29 19:27:04 2001), 站内信件


that would be a much more interesting topic.
pjones, I guess an AI doesn't have to be a non-deterministic program.
That is, an AI program doesn't have to rely on things like
random() to generate non-deterministic(unpredictable is more
appropriate a word?) results.  An AI program could be deterministic, and
yet produce unpredictable results.  Imagine a program implementing
a very very complex set of rule, and taking a very very complex set
of input.  They are so complex that the operator can not have total
control over which exact set of input to feed to the program, nor
how to choose a particular set of input to get a particularly expected result.
Therefore, the output will be unpredictable.

--
他号令便号令好了,又何必安静?

※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 208.243.29.169]
发信人: qili (qili), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Mon Jan 29 20:07:24 2001), 站内信件

one important treat of human thinking is our ability to
a) update knowledge/rules by which we deduce; and
b) deduce under incomplete and inperfect information; and
c) select a subset of knowledge/rules to work with (fuzziness?)

I tend to think that unless machinese are taught to work
*unexact* information, they will remain machines.

【 在 magicfat (魔法胖子~missing my Friend) 的大作中提到: 】
: that would be a much more interesting topic.
: pjones, I guess an AI doesn't have to be a non-deterministic program.
: That is, an AI program doesn't have to rely on things like
: random() to generate non-deterministic(unpredictable is more
: appropriate a word?) results.  An AI program could be deterministic, and
: yet produce unpredictable results.  Imagine a program implementing
: a very very complex set of rule, and taking a very very complex set
: of input.  They are so complex that the operator can not have total
: control over which exact set of input to feed to the program, nor
: how to choose a particular set of input to get a particularly expected result.
: Therefore, the output will be unpredictable.


--
※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 32.101.37.70]
发信人: microbe (奇妙的微生物), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Mon Jan 29 20:12:49 2001), 站内信件


【 在 magicfat (魔法胖子~missing my Friend) 的大作中提到: 】
: that would be a much more interesting topic.
: pjones, I guess an AI doesn't have to be a non-deterministic program.

If course not, who said that?  :-)

I doubt there is any accurate definition of AI.  One definition I heard
in the AI class was "AI is what AI researchers are working on".

: That is, an AI program doesn't have to rely on things like
: random() to generate non-deterministic(unpredictable is more
: appropriate a word?) results.  An AI program could be deterministic, and
: yet produce unpredictable results.  Imagine a program implementing
: a very very complex set of rule, and taking a very very complex set
: of input.  They are so complex that the operator can not have total
: control over which exact set of input to feed to the program, nor
: how to choose a particular set of input to get a particularly expected result.
: Therefore, the output will be unpredictable.


--
我在门外坐了两天两夜,看着天空在不断的变化。
我才发现,虽然我到这里很久,却从来没有看清楚这片沙漠。

※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 160.39.200.28]
发信人: microbe (奇妙的微生物), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Mon Jan 29 20:14:55 2001), 站内信件

【 在 qili (qili) 的大作中提到: 】
: one important treat of human thinking is our ability to
: a) update knowledge/rules by which we deduce; and
: b) deduce under incomplete and inperfect information; and
: c) select a subset of knowledge/rules to work with (fuzziness?)
: I tend to think that unless machinese are taught to work
: *unexact* information, they will remain machines.

What is *unexact* information?

: 【 在 magicfat (魔法胖子~missing my Friend) 的大作中提到: 】
: : that would be a much more interesting topic.
: : pjones, I guess an AI doesn't have to be a non-deterministic program.
: : That is, an AI program doesn't have to rely on things like
: : random() to generate non-deterministic(unpredictable is more
: : appropriate a word?) results.  An AI program could be deterministic, and
: : yet produce unpredictable results.  Imagine a program implementing
: : a very very complex set of rule, and taking a very very complex set
: : of input.  They are so complex that the operator can not have total
: : control over which exact set of input to feed to the program, nor
: : how to choose a particular set of input to get a particularly expected result.
: : Therefore, the output will be unpredictable.


--
我在门外坐了两天两夜,看着天空在不断的变化。
我才发现,虽然我到这里很久,却从来没有看清楚这片沙漠。

※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 160.39.200.28]
发信人: qili (qili), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Mon Jan 29 20:49:11 2001), 站内信件

【 在 microbe (奇妙的微生物) 的大作中提到: 】
: 【 在 qili (qili) 的大作中提到: 】
: : one important treat of human thinking is our ability to
: : a) update knowledge/rules by which we deduce; and
: : b) deduce under incomplete and inperfect information; and
: : c) select a subset of knowledge/rules to work with (fuzziness?)
: : I tend to think that unless machinese are taught to work
: : *unexact* information, they will remain machines.
: What is *unexact* information?

I actually struggled to find an exact word to describe it, :)

think about how you would recognize a picture of a friend of yours.
you looked at the picture quickly, without getting precise information
(thus, you do not know exactly the color, brightness, inter-pixel
relationships). Yet, you can tell quickly if the picture is of
your friend.

today's computers don't work that way. They want precise information
and every piece of information about that picture. Worse yet,
having scrathes, or marks on that picture, or taking a picture of the same person
from a slightly different angle may throw that computer off.

The point here is that our computers today require precise
information to work on.

From that point of view, my dog is probably smarter than all of the
world's computers combined, :)

--
※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 32.101.37.220]
发信人: dudu (神奇主), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Mon Jan 29 21:03:43 2001) WWW-POST

What I learned from courses is
AI is to think like humans/think rationally/act like
humans/act rationally.

I feel the important features in AI are
able to learn
uncertainty
reasoning

【 在 microbe (奇妙的微生物) 的大作中提到: 】
:
: 【 在 magicfat (魔法胖子~missing my Friend) 的大作中提到: 】
: : that would be a much more interesting topic.
: : pjones, I guess an AI doesn't have to be a non-deterministic program.
:
: If course not, who said that?  :-)
:
: I doubt there is any accurate definition of AI.  One definition I heard
: in the AI class was "AI is what AI researchers are working on".
:
: : That is, an AI program doesn't have to rely on things like
: : random() to generate non-deterministic(unpredictable is more
: : appropriate a word?) results.  An AI program could be deterministic, and
: : yet produce unpredictable results.  Imagine a program implementing
: : a very very complex set of rule, and taking a very very complex set
: : of input.  They are so complex that the operator can not have total
: : control over which exact set of input to feed to the program, nor
: : how to choose a particular set of input to get a particularly expected result.
: : Therefore, the output will be unpredictable.
:
:


--
※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: rb16v-93-229.re]
发信人: microbe (奇妙的微生物), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Mon Jan 29 21:26:25 2001), 站内信件


That is called "pattern matching".  That's not "unexact information".
That's just paralell processing of exact information.  It's fast because
it's a paralell process.  We have about 100 billion neutrons, each being
a very slow CPU (1000 ops/sec), but very fast as a whole.  It's fundamentally
different from how computers work (very few very fast CPUs).

【 在 qili (qili) 的大作中提到: 】
: 【 在 microbe (奇妙的微生物) 的大作中提到: 】
: : What is *unexact* information?
: I actually struggled to find an exact word to describe it, :)
: think about how you would recognize a picture of a friend of yours.
: you looked at the picture quickly, without getting precise information
: (thus, you do not know exactly the color, brightness, inter-pixel
: relationships). Yet, you can tell quickly if the picture is of
: your friend.
: today's computers don't work that way. They want precise information
: and every piece of information about that picture. Worse yet,
: having scrathes, or marks on that picture, or taking a picture of the same person
: from a slightly different angle may throw that computer off.
: The point here is that our computers today require precise
: information to work on.
: From that point of view, my dog is probably smarter than all of the
: world's computers combined, :)


--
我在门外坐了两天两夜,看着天空在不断的变化。
我才发现,虽然我到这里很久,却从来没有看清楚这片沙漠。

※ 修改:.microbe 于 Jan 29 21:27:39 修改本文.[FROM: 160.39.200.28]
※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 160.39.200.28]
发信人: whh (大不了爱吃爱吃), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Mon Jan 29 22:04:40 2001), 站内信件


【 在 qili (qili) 的大作中提到: 】
: today's computers don't work that way. They want precise information
: and every piece of information about that picture. Worse yet,
: having scrathes, or marks on that picture, or taking a picture of the same person
: from a slightly different angle may throw that computer off.
: The point here is that our computers today require precise
: information to work on.

No, the problem is we still don't know how human beings do the recognition.
If we do, we can certainly invent an algorithm simulating what human beings
do. Computers and human beings take different ways to make decision on
the same input, but the later is superior, at least now.

: From that point of view, my dog is probably smarter than all of the
: world's computers combined, :)


--
※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 128.107.148.80]
发信人: microbe (奇妙的微生物), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Mon Jan 29 22:17:17 2001), 站内信件


【 在 whh (大不了爱吃爱吃) 的大作中提到: 】
: 【 在 qili (qili) 的大作中提到: 】
: : today's computers don't work that way. They want precise information
: : and every piece of information about that picture. Worse yet,
: : having scrathes, or marks on that picture, or taking a picture of the same person
: : from a slightly different angle may throw that computer off.
: : The point here is that our computers today require precise
: : information to work on.
: No, the problem is we still don't know how human beings do the recognition.
: If we do, we can certainly invent an algorithm simulating what human beings
: do. Computers and human beings take different ways to make decision on
: the same input, but the later is superior, at least now.

The later is not necessarily superior.  It all depends on the nature of
problem.  Some problems are easier to solve by the computer way, some
are easier by the human way, and the others can be solved equally well.

: : From that point of view, my dog is probably smarter than all of the
: : world's computers combined, :)


--
我在门外坐了两天两夜,看着天空在不断的变化。
我才发现,虽然我到这里很久,却从来没有看清楚这片沙漠。

※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 160.39.200.28]
发信人: qili (qili), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Tue Jan 30 09:29:57 2001), 站内信件

【 在 microbe (奇妙的微生物) 的大作中提到: 】
: That is called "pattern matching".  That's not "unexact information".
: That's just paralell processing of exact information.  It's fast because
: it's a paralell process.  We have about 100 billion neutrons, each being
: a very slow CPU (1000 ops/sec), but very fast as a whole.  It's fundamentally
: different from how computers work (very few very fast CPUs).

the point I wanted to get cross is that human brains are very good
at proximating: getting 80% of the answer right quickly, while computers
are good at working with precise information.

I don't see our computers behave like human brains unless we understand how
human brains work first. Before that happens, I am reluctant to state that
we can simulate human brains with computers.

--
※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 206.189.24.11]
发信人: magicfat (魔法胖子~missing my Friend), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Tue Jan 30 10:27:42 2001), 站内信件

【 在 microbe (奇妙的微生物) 的大作中提到: 】
: 【 在 magicfat (魔法胖子~missing my Friend) 的大作中提到: 】
: : that would be a much more interesting topic.
: : pjones, I guess an AI doesn't have to be a non-deterministic program.
: If course not, who said that?  :-)

pjones said that! pjones said that! 8-)

: I doubt there is any accurate definition of AI.  One definition I heard
: in the AI class was "AI is what AI researchers are working on".

Well I'd rather stay with Turing's -- if you can't tell if it's a
human being or a machine by talking to it only, it is an AI.

--
他号令便号令好了,又何必安静?

※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 208.243.29.169]
发信人: magicfat (魔法胖子~missing my Friend), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Tue Jan 30 10:37:22 2001), 站内信件

【 在 qili (qili) 的大作中提到: 】
: the point I wanted to get cross is that human brains are very good
: at proximating: getting 80% of the answer right quickly, while computers
: are good at working with precise information.
: I don't see our computers behave like human brains unless we understand how
: human brains work first. Before that happens, I am reluctant to state that
: we can simulate human brains with computers.

I would have to disagree with you on this.  I'm sure that knowing
how human brain works would definitely help us develop better AI.
However, that is not a must-have(anybody reminds me the mathematical term? 8-)
If we stick to Turing's definition of AI, a machine can be implemented
in a manner that is entirely different from the work scheme of human
brain, yet it is still an AI, if it can achieve a comparatively equivalent
level of human intelligence.

One example, when you play chess with Deep Blue, you wouldn't know
it is a machine from its chess playing skills (you'll be able to tell
maybe after you get used to its style, but that's a different matter)

--
他号令便号令好了,又何必安静?

※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 208.243.29.169]
发信人: qili (qili), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Tue Jan 30 10:52:53 2001), 站内信件

【 在 magicfat (魔法胖子~missing my Friend) 的大作中提到: 】
: 【 在 qili (qili) 的大作中提到: 】
: : I don't see our computers behave like human brains unless we understand how
: : human brains work first. Before that happens, I am reluctant to state that
: : we can simulate human brains with computers.
: I would have to disagree with you on this.  I'm sure that knowing
: how human brain works would definitely help us develop better AI.
: However, that is not a must-have(anybody reminds me the mathematical term? 8-)
: If we stick to Turing's definition of AI, a machine can be implemented
: in a manner that is entirely different from the work scheme of human
: brain, yet it is still an AI, if it can achieve a comparatively equivalent
: level of human intelligence.

I will concede that it is possible, maybe even probable, to get a machine
to behave like a human without understanding how human brains work. For example,
machines do 1+1=2 as well as a human brains.

So if you narrowly interpret Touring, some of the machines already has
AI. However, I think his point is more on broad/broader terms like if
the machine would behave / react to a vast array of tasks / inputs like
we do.

While we can make specialized machines to react to a specific task
in a human way (like Deep Blue), we still don't have a machine that
can take on daily jobs like a human (Hal would be a good example).


--
※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 129.79.6.176]
发信人: magicfat (魔法胖子~missing my Friend), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Tue Jan 30 12:10:37 2001), 站内信件

【 在 qili (qili) 的大作中提到: 】
: I will concede that it is possible, maybe even probable, to get a machine
: to behave like a human without understanding how human brains work. For example,
: machines do 1+1=2 as well as a human brains.
: So if you narrowly interpret Touring, some of the machines already has
: AI. However, I think his point is more on broad/broader terms like if
: the machine would behave / react to a vast array of tasks / inputs like
: we do.
: While we can make specialized machines to react to a specific task
: in a human way (like Deep Blue), we still don't have a machine that
: can take on daily jobs like a human (Hal would be a good example).

For that I agree, but, hey, we are talking about what is AI, not whether
we have AI right now, nor whether we can ever produce some "true" AI
someday.  Besides, I think we are moving on the right direction.
In some areas we are just confined by the hardware capacity, rather than
AI theories.

--
他号令便号令好了,又何必安静?

※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 208.243.29.169]
发信人: qili (qili), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Tue Jan 30 12:37:24 2001), 站内信件

【 在 magicfat (魔法胖子~missing my Friend) 的大作中提到: 】
: For that I agree, but, hey, we are talking about what is AI, not whether
: we have AI right now, nor whether we can ever produce some "true" AI
: someday.  Besides, I think we are moving on the right direction.
: In some areas we are just confined by the hardware capacity, rather than
: AI theories.

well, a more fundamental question, before we take that giant leap of
faith, is "do we want to have machines that are capable of having
intelligence?" Robotic ethics probably would come into play here,
more so than technical feasibility.

--
※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 206.189.24.6]
发信人: microbe (奇妙的微生物), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Tue Jan 30 13:11:21 2001), 站内信件


【 在 magicfat (魔法胖子~missing my Friend) 的大作中提到: 】
: 【 在 qili (qili) 的大作中提到: 】
: : the point I wanted to get cross is that human brains are very good
: : at proximating: getting 80% of the answer right quickly, while computers
: : are good at working with precise information.
: : I don't see our computers behave like human brains unless we understand how
: : human brains work first. Before that happens, I am reluctant to state that
: : we can simulate human brains with computers.
: I would have to disagree with you on this.  I'm sure that knowing
: how human brain works would definitely help us develop better AI.
: However, that is not a must-have(anybody reminds me the mathematical term? 8-)

necessary condition? hehe..

: If we stick to Turing's definition of AI, a machine can be implemented
: in a manner that is entirely different from the work scheme of human
: brain, yet it is still an AI, if it can achieve a comparatively equivalent
: level of human intelligence.
: One example, when you play chess with Deep Blue, you wouldn't know
: it is a machine from its chess playing skills (you'll be able to tell
: maybe after you get used to its style, but that's a different matter)

right.  chess is one thing that can be solved equally well by computer (serial
search) and human (patten matching).

--
我在门外坐了两天两夜,看着天空在不断的变化。
我才发现,虽然我到这里很久,却从来没有看清楚这片沙漠。

※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 160.39.200.28]
发信人: whh (大不了爱吃爱吃), 信区: ITnews
标  题: Re: Well, what is AI?
发信站: The unknown SPACE (Tue Jan 30 17:46:27 2001), 站内信件

Well, by then, we might already have ways to re-gene ourselves and
make us smarter than AI we generate. Even if AI is smarter than us,
it is not a bad idea since Darwin's theory gets in here. Why should
we maintain an inferior control over the universe that clearly does
not belong to us? Who knows human beings were not generated by some
other species having lower intelligence (who might have been killed by us)?

【 在 qili (qili) 的大作中提到: 】
: 【 在 magicfat (魔法胖子~missing my Friend) 的大作中提到: 】
: : For that I agree, but, hey, we are talking about what is AI, not whether
: : we have AI right now, nor whether we can ever produce some "true" AI
: : someday.  Besides, I think we are moving on the right direction.
: : In some areas we are just confined by the hardware capacity, rather than
: : AI theories.
: well, a more fundamental question, before we take that giant leap of
: faith, is "do we want to have machines that are capable of having
: intelligence?" Robotic ethics probably would come into play here,
: more so than technical feasibility.


--
※ 来源:.The unknown SPACE bbs.mit.edu.[FROM: 128.107.148.80]

[返回]
赞助链接
未名交友
将您的链接放在这儿
 

Site Map - Contact Us - Terms and Conditions - Privacy Policy

版权所有,未名空间(mitbbs.com),since 1996